Leveraging GNN-Driven Inference with Pub/Sub Messaging for Coherent Data Erasure in Multi-Node Cloud Systems

Large-scale, distributed data storage is made possible by cloud computing, which also improves system performance and availability. However, a recurring and unsolved problem is making sure that data is consistently and completely deleted across several servers. Copies of a file that has been deleted by a server might still be present on other network nodes, which could result in data leakage, violations of privacy laws, and wasteful storage resource usage. In order to resolve the discrepancy, this work presents an intelligent system that combines Graph Neural Networks (GNNs) with a Publish/Subscribe (Pub/Sub) messaging to make this deletion. The entire storage infrastructure is modeled as a dynamic graph, where servers and data blocks are represented as nodes, and replication relationships as edges, enhanced with attributes like last access time, trust levels, and file size. When data deletion takes place, the Pub/Sub model notifies all subscribed nodes, which then use GNN to predict other possible data locations. Based on the prediction confidence, the system requests admin approval, or deletes the data automatically. This approach is highly scalable, and reduces duplicate actions and helps in proper detection of the deleted data from other servers. Integrated auditing feature which is similar to blockchain, provides a secure and record of all deletion operations, increasing liability. The proposed framework enables reliable, efficient, which is really crucial in data-sensitive sectors such as healthcare, finance, and education. It's design supports both operational efficiency and adherence to privacy regulations across distributed storage environments. Thus this approach has been taken.